a Centro de Biología Molecular "Severo Ochoa", CSIC/Universidad Autónoma de Madrid, C/Nicolás Cabrera 1, Universidad Autónoma, Cantoblanco, 28049 Madrid, Spain.b Experimental Therapeutics and Translational Oncology Program, Instituto de Biología Molecular y Celular del Cáncer, CSIC/ Universidad de Salamanca, Campus M. de Unamuno s/n, 37007-Salamanca, Spain.
* Corresponding author at: Centro de Biología Molecular "Severo Ochoa", CSIC-Universidad Autónoma de Madrid, C/Nicolás Cabrera 1, Universidad Autónoma, Cantoblanco, 28049 Madrid, Spain. Phone: +34-923-294672; Fax: +34-923-294813. E-mail address: ccobaleda@cbm.uam.es (Cesar Cobaleda)
September 2011
Summary
The research fields of developmental biology and oncology have always been tightly linked, since the times of Rudolf Virchow's cellular theory ("omnis cellula e cellula") and embryonal rest hypothesis. On the other side, for many years, contemporary cancer research has been mainly focused on the altered controls of proliferation in tumoral cells. This has been reflected in the therapeutic approaches employed in the clinic to treat the patients: with very few exceptions, anti-cancer treatments are targeted at the mechanisms of abnormal tumoral growth. Such therapies, however, are very unspecific, highly toxic and, ultimately, inefficient in most cases. In the last years, a new recognition of the role of aberrant differentiation at the root of cancer has arisen, mainly driven by the coming of age of the "cancer stem cell" (CSC) theory. From this point of view, the comprehensive knowledge of the developmental mechanisms by which normal cells acquire their identity is essential to understand how these controls are deregulated in tumours. New insights into the mechanisms that maintain the molecular boundaries of cell identity have been gained from the study of induced pluripotency, showing that cell fate can be much more susceptible to change than previously thought. Applied to cancer, these findings imply that the oncogenic events that take place in an otherwise healthy cell lead to a reprogramming of the normal cellular fate and establish a new pathologic developmental program. Therefore, cancer reprogramming and cellular plasticity are closely related, since only some cells possess the plasticity required to allow reprogramming to occur, and only some oncogenic events can, in the right plastic cell, induce this change. Here we discuss the latest findings in the fields of cellular plasticity and reprogramming and we consider their consequences for our understanding of cancer development and treatment.
Historical perspective
The search for the capacity of regenerating disease-affected organs is probably as old as mankind (Odelberg, 2004). The examples are abundant in ancient religions, from the Egyptian god Osiris, who resurrected and recomposed his maimed body from the pieces that had been thrown into the Nile, to the legendary Hydra that could regenerate its severed heads. Or the mythological Prometheus, who had his viscera eaten by an eagle every day, only to regenerate them again. But also from a more scientific point of view, it was already noticed by Aristotle (384-322 BC) that lizards can regenerate their tails after amputation. But until the 18th century this knowledge was mainly anecdotic, and only with the arrival of the Age of Enlightenment, regeneration and plasticity will become the matter of scientific research. In 1712, Réaumur describes the regeneration of the limbs and claws of crayfish (Réaumur, 1712); in 1744, Trembley discovers that a part of the Hydra polyp can regenerate the complete organism (Trembley, 1744); in 1769, Spallanzani reports that tadpoles can regenerate their tails and salamanders their amputated jaws, limbs and tails (Spallanzani, 1769). The research performed during most of the 19th and first half of the 20th centuries showed that, for regeneration to occur, the cells that are normally forming part of the organs are not sufficient, and a special type of cells are required: the progenitor cells (Odelberg, 2004; Birnbaum and Sánchez Alvarado, 2008). The origin of these cells was not very clear (and it is still a matter of debate and intense research, in fact, see (Sánchez Alvarado, 2000; Kragl et al., 2009; Rinkevich et al., 2011)); for some tissues, like skin, blood, muscles or bones, progenitors were shown to exist in the tissues in small numbers, and to become activated as a consequence of the lesions. In other cases, the progenitors seemed to arise from mature cells that become dedifferentiated. The best example supporting this possibility has been described in primitive vertebrates like the urodeles (e.g. salamanders and axolotls). In these animals, after a wound harms the organism, the cells from the normal tissues form a group of cells known as the regenerative blastema, which will generate all the tissues in the new limb/tail (Chalkey, 1954; Bodemer and Everett, 1959; Hay and Fischman, 1961). It has long been held that the blastema was the result of cellular dedifferentiation to progenitors. However, the most recent findings seem to indicate that there is no cellular dedifferentiation to progenitors involved in this process, and the regeneration is always due to the action of resident tissue-specific stem cells and progenitors, thus questioning the role of mature cellular plasticity in tissue regeneration (Kragl et al., 2009; Rinkevich et al., 2011). We have therefore seen how the study of "naturally" occurring regeneration opened the way to a new understanding of the stem cell-based architecture of the organs and tissues, especially with the study of primitive vertebrates. In 1952, amphibians also provided the first animal model of experimentally-induced reprogramming when Briggs and King generated Xenopus tadpoles by transplanting the nucleus of cells from the blastula into oocytes, therefore reverting the cellular differentiation program (Briggs and King, 1952). Afterwards, it was shown that more differentiated cells, like those from the intestinal epithelia, could also be reprogrammed by nuclear transfer (Gurdon, 1962). These landmark findings undoubtedly showed that the genetic potential of cells was not lost during differentiation, and that development did not imply genetic changes. This principle was extended to mammals with the cloning of Dolly the sheep in 1997 (Wilmut et al., 1997). This was the ultimate proof showing that the changes that occur during differentiation are totally reversible, and demonstrated that the fate restrictions that take place during development are the result of epigenetic modifications. These studies also showed that there were factors in the oocyte cytoplasm capable of inducing a reprogramming that led to the appearance of a totipotent phenotype. In a parallel way, the search for the molecular regulators responsible for establishing and controlling cellular identity led finally to the identification of the factors capable of reprogramming cellular fate. In 1987, it was shown that ectopic expression of the Antennapedia homeotic gene lead to changes in the body plan of Drosophila, that got extra legs instead of antennae (Schneuwly et al., 1987). Also, Gehring et al. showed that the ectopic expression of eyeless controlled eye development and led to the development of ectopic eyes in the fly's legs (Gehring, 1996). In mammals, the first master regulator factor to be identified was MyoD, which was shown to be capable of transdifferentiating fibroblasts into the myogenic lineage (Davis et al., 1987). Other examples of factors with fate-reprogramming capacity are C/EBPα, capable of mediating the transdifferentiation of mouse B cells into macrophages (Xie et al., 2004) or Pax5, whose loss leads to the dedifferentiation of committed B cells (Nutt et al., 1999; Cobaleda et al., 2007a; Cobaleda and Busslinger, 2008). All these data proved that the lack or excess of just one factor could lead to a radical alteration of the transcriptional profile and could cause stable fate changes. This evidence, together with the one coming from reprogramming by nuclear transplantation, paved the way to the search for the factors capable of reprogramming to full pluripotency that led, in 2006, to the identification of the four transcription factors capable of inducing pluripotency in virtually every kind of terminally differentiated cells (Takahashi and Yamanaka, 2006). We will discuss this aspect with more detail afterwards. On the other side, cancer has also been recognized as a distinct pathological entity since the origins of mankind. The first references are the Edwin Smith and Ebers papyri from the 1600 BC and 1500 BC, approximately (Hajdu, 2004). The Edwin Smith papyrus contains the first mention and description of breast cancer, and it concludes that there is no treatment for the disease. Cancer was not so common in ancient times, mainly because life span was much shorter, but it was already clearly recognized. Hippocrates (460-375 BC) realized that growing tumors occurred typically in adults and they reminded him of a moving crab, which led to the terms carcinos and cancer. Celsus (25 BC-AD 50) also compared cancer with a crab, because it penetrates the surrounding organs like if it had claws; Celsus introduced the first classification for breast cancer and advocated for surgical therapy. Furthermore, he already realized that tumors could only be cured if they were removed in their early stages and that, even after removal and wound healing, breast carcinomas tended to recur causing swelling in the armpit and, finally, death by spreading throughout the body. Galen (131-AD 200) already recommended surgery by cutting a wide margin of healthy tissue around the edges of the tumor (Hajdu, 2004). If we jump now to our days, it seems disappointing to see how little those old critical findings have been overcome by modern medicine, 2000 years later. Indeed, still today, clean surgical margins and lack of lymph node invasion are the most important prognostic markers for the successful eradication of solid tumors, and only if tumors are completely resected before they metastasize (something that it is anyhow impossible to determine with current technologies) can curation be guaranteed. However, in the last thirty years we have gained an enormous knowledge about the molecular biology of the disease. In 1979, it was shown that the phenotype of transformed cells could be transferred to normal fibroblasts by DNA transfection (Shih et al., 1979), a finding that lead to the rapid molecular cloning of the first human oncogene (the RAS gene), simultaneously by several groups (Goldfarb et al., 1982; Lane et al., 1982; Parada et al., 1982; Santos et al., 1982). Since then, many genes have been described as being either oncogenes or tumor suppressors, and the molecular mechanisms of their transforming capabilities have been analyzed to great detail, in close relationship with their functions in "normal" conditions. This is a field that has expanded tremendously in the last decades, and a comprehensive study of the topic falls out of the scope of this revision. However, there are some aspects that must be taken into account for posterior debate. A very important one is the fact that, for many types of tumors, specific genetic mutations have been shown to correlate closely with the phenotype of the tumors, suggesting that the oncogenic alterations might be acting as new specification factors that determine the tumor appearance and/or phenotype. This association is especially evident in the case of mesenchymal tumors caused by chromosomal aberrations (Sánchez-García, 1997; Cobaleda et al., 1998). In 2000, Hanahan and Weinberg summarized the main features that had to be disrupted in normal cellular behavior in order for allow a tumor to appear and progress (Hanahan and Weinberg, 2000), and this list has expanded with the years (Hanahan and Weinberg, 2011). These main aspects are related with the survival and proliferation of cancer cells, but it must be noted that most of them are equally shared by non-malignant tumors (Lazebnik, 2010). However, all the aspects related to the alterations of the normal developmental regulatory mechanisms in tumorigenesis have received much less attention. But in fact, if cellular fate was carved into stone, cancer would be impossible, since no new lineages could be generated other than the normal, physiologic ones. Here is where the normal mechanisms regulating cellular identity and plasticity play an essential role in allowing cancers to arise and hopefully, as we will discuss, they might be the key to its eradication. The specification of cellular identity during development and differentiation is a dynamic process that starts with stem and progenitor cells and ends with terminal differentiation into each specialized cellular type. In this progression there can be many cellular intermediates; some of them are transient, and some can be long-lasting, but the maintenance of cellular identity at each stage is determined by the signals from the environment and, in an intrinsic manner, by specific transcription factors and epigenetic modifiers that establish a defined chromatin architecture and a specific gene expression profile. As we have seen, evidences about cellular plasticity had being accumulating for decades (Hochedlinger and Jaenisch, 2006; Gurdon and Melton, 2008; Graf and Enver, 2009; Vicente-Dueñas et al., 2009a), but the latest findings in the field of reprogramming have definitively shown how switching to a different phenotype can be a lot easier than previously expected, and can have real physiological relevance, beyond basic research. Cancer is a perfect example of pathological reprogramming in which, from a normal tissue, a whole new differentiation lineage is opened with its own hierarchy and structure (Reya et al., 2001; Sánchez-García et al., 2007). So, without forgetting the so well-studied aberrant proliferation, reprogramming is an essential part of the tumorigenesis process, and it is closely dependent on the cellular plasticity of the cancer-initiating cells. The term plasticity, as we will use it here, refers to the ability of cells (stem or differentiated) to adopt the biological properties (gene expression profile, phenotype, etc.) of other differentiated types of cells (belonging to the same or different lineages). This definition comprises also the property of competence, i.e. the ability of stem cells and progenitors to give rise to their different descendant lineages during normal development. We use such an ample definition of the term precisely to reflect the fact that the molecular mechanisms that are important for progenitors' competence during normal development are the same ones responsible for the plasticity changes of more differentiated types of cells, both in pathological processes and in experimentally-induced reprogramming. Here we will discuss the vital role of cellular plasticity in the origin and maintenance of tumoral cells. We will first revise the latest research discoveries in the fields of normal developmental and experimentally-induced plasticity, and afterwards will link these findings with what we know about cancer biology.
Lineage commitment and cellular identity
Adult stem cells are the responsible of generating all the different specialized cellular types forming the organism. The majority of them perform this job throughout the whole life of the organism, thanks to their self-renewal capacity. This property allows them to divide asymmetrically, therefore given rise to a new identical daughter stem cell and to a multipotential progenitor, lacking self-renewal capacity, which will give rise to all the differentiated tissue cells. Although it is known that there are some specific factors that are essential for the specification and maintenance of stem cell identity (Boyer et al., 2005), the molecular bases of the choice that stem cells have to make between maintaining competence (i.e. plasticity) or entering into the differentiation programs are not yet completely understood (Niakan et al., 2010). In this context, a first important aspect to consider is the fact that the stem cell population itself is intrinsically heterogeneous. This means that the "stemness" is not a static condition defined by stable, constant levels of expression of intrinsic stem factors and surface stem markers, but it is more of a continuum that moves within certain margins. For example, in a clonal population of haematopoietic progenitor cells, there is a Gaussian distribution of the levels of expression of Sca-1, one of the most classical stem cell markers (Chang et al., 2008). Furthermore, cells at both the low- or high-end levels of expression can, when isolated, regenerate the whole population with all the range of expression levels. However, every one of these sub-populations, defined by their levels of a surface marker, also expresses different transcriptomes, and has therefore distinct intrinsic differentiation tendencies towards different lineages. Therefore, each individual cell in the stem population represents a metastable transitional point in a continuum of constantly changing transcriptomes. In fact, this is most probably the mechanism at the basis of the stochastic choice of lineage, when some cells approach too much to the "edges" of the normal distribution and the transcriptome changes become irreversible (Chang et al., 2008). In 1957 Waddington conceptualized the irreversible process of cellular differentiation as marbles falling down a slope (Waddington, 1957). This metaphorical concept has regained new momentum with the mathematical interpretation of transcriptional cellular states as Gene Regulatory Networks (GRNs). In this type of analysis, pluripotency is represented as a mathematical attractor (a condition towards which a dynamical system tends to progress over time), in such a way that the points (cells) that get close enough to the attractor remain close even if slightly disturbed. This attractor is surrounded by a "differentiation landscape" where other stable cellular fates are represented by stable "valleys" and differentiation routes towards them are "channels" through which the cells move (Enver et al., 2009; Huang, 2009). Under this light, pluripotency can be considered as a dynamic state of controlled heterogeneity within a population, where small individual fluctuations in the levels of expression of transcription factors and epigenetic regulators maintain a global status of apparent stability. The cells that approach the limits of the attractor (those who, in their random fluctuations, go too far from the middle point of the Gaussian curve) are therefore more prone to differentiate, suggesting that commitment, although rare, is an spontaneous phenomenon (unless it is specifically triggered by an external signal that unbalances the dynamic equilibrium) (Huang, 2009).
Maintenance of cellular identity throughout the differenciation process
Although in some rare cases they are unipotent (e.g. spermatogonial stem cells), adult stem cells are usually multipotent, and they can give rise to a wide range of differentiated cell types. In the first instance, stem cells lose their self-renewal potential (their stemness) and start the differentiation process by becoming multipotential progenitors. We have seen that the differentiation program can be pre-set already by the oscillatory patterns of gene expression at the stem cell population level, and that cells lying at the different ends of specific gradients of gene expression can have opposite differentiation preferences (Chang et al., 2008). So, once they leave the stem cell state, the cells start making lineage choices that are usually mutually excluding and are normally conceptualized in a branching pattern. These alternative options are usually controlled by the cross-antagonism between transcription factors with competing, opposing functions (Swiers et al., 2006; Loose et al., 2007). A very well characterized developmental system is hematopoietic differentiation where several models of lineage-specification have been identified which seem to be based on the aforementioned mechanism. For example, the choice between erythroid/megakaryocyte or myeloid-monocytic fates at the level of erythromyeloid progenitors is controlled by the reciprocal inhibition between the transcription factors GATA-1 and PU.1, therefore creating a binary decision for the progenitor (Laiosa et al., 2006; Enver et al., 2009). The bipotent progenitor itself would therefore be this intermediate state created and maintained by the equilibrium between the both factors. This fact helps understanding the phenomenon of multilineage gene priming, in which uncommitted progenitors present low levels of simultaneous expression of multiple transcription factors corresponding to different mature cell types and possessing antagonistic functions (Hu et al., 1997; Enver et al., 2009). In general, there seems to be a progressive loss of developmental potential in a hierarchical process that moves through sequential differentiation options and in which, at any given point, a progenitor would only have to choose between two mutually exclusive options (Brown et al., 2007; Ceredig et al., 2009). Additionally, in the process of maturation into a given lineage, the progenitors will receive (and react to) the necessary extrinsic signals (for example, cytokines) that, according to this model, would be more permissive than instructive.
Maintenance of the cellular identity of mature differentiated cells
Plasticity, in normal development, is a property that is "intended" to be restricted to stem cells and progenitors. In general, the final differentiated cellular types of any given organ or tissue possesses stable identities, in consequence with the fact that they usually are highly specialized cells with very specific physiological functions. Therefore, it would not make sense, from the biological point of view, that a specialized cell would be the source of other differentiated cell types. This, as we have mentioned, is the role of stem cells, with their physiological plasticity (i.e., normal competence) that we have previously discussed. However, the concept of the stability of differentiated cell types has been shaken by the discovery of the fact that the 4 Yamanaka transcription factors (4Y TFs) Oct4, Sox2, c-Myc and Klf4 (Takahashi and Yamanaka, 2006) are enough for the reprogramming of most differentiated cells types into induced pluripotent stem cells (iPSCs). This finding has altered our notion of the latent developmental potential hidden in differentiated cells, showing how it can be "awakened" by experimental manipulations in the laboratory. This, as we have described, was already known to a certain extent from the nuclear reprogramming experiments performed in amphibians more than 50 years ago (Briggs and King, 1952; Gurdon, 1962). Nevertheless, although those experiments already proved that the cell nucleus could be reprogrammed from a differentiated cell type into a pluripotent progenitor, Yamanaka's experiments showed that only 4 factors were actually enough to make the whole process possible. We have seen that, a more modest level, it had already been proven that the overexpression or loss of individual transcription factors could induce fate changes in differentiated cells (MyoD, C/EBPa, Pax5, etc). Although these were examples of transdifferentiation taking place between closely related cell types, they already pointed the way for the search of the factors capable of reprogramming to full pluripotency. Since the differentiated state is the more stable one (indicating that the GRNs are less subject to fluctuation), where the cells have reached after "rolling down" the differentiation pathway in the normal process of development, therefore an "activation energy" is required to move the cells "uphill" to become again pluripotent. Conceptually, there are at least two main possible scenarios to explain the population dynamics in the process of reprogramming to pluripotency (Yamanaka, 2009): one possibility (the so-called elite model) is that only some cells can be reprogrammed, and these are the ones that are selected among the entire population, since they are the only ones that are receptive to the action of the reprogramming factors. Alternatively, it might happen that all the differentiated cells are equally capable of undergoing reprogramming, and it is only due to technical or methodological reasons that we are not able to reveal this potential in all of them (stochastic model). According to the accumulating evidences, it would seem that the stochastic model is the one that is closer to reality and that, given the right combination of factors; any cell could be reprogrammed to pluripotency (Yamanaka, 2009). However, as we have mentioned, this is a developmentally and energetically unfavourable process, a fact that is evidenced by several details. The most obvious one is the very low efficiency of the reprogramming process, even in the most favourable conditions. This fact clearly indicates that, independently on how many cells of the population are initially responsive to the reprogramming factors, very few of them can complete the path towards full reprogramming (Yamanaka, 2009). Also, this is a gradual process in which several non-physiological cellular intermediates can be isolated (Mikkelsen et al., 2008; Stadtfeld et al., 2008). The study of these incompletely reprogrammed intermediates has revealed that they have re-activated the self-renewal and maintenance stem cell genes, but not yet those of pluripotency; also, these stages of aborted reprogramming have not been able to completely repress the expression of lineage-specific transcription factors and retain persistent DNA hypermethylation marks as a proof of their failure in achieving complete epigenetic remodelling (Mikkelsen et al., 2008). But perhaps the most patent proof of the difficulty of the process of full reprogramming to pluripotency is the persistence of an epigenetic memory in the iPCs that makes them more prone to re-differentiate into the lineages from which they were initially derived, indicating that a complete elimination of the initial epigenetic program cannot yet be achieved (Kim et al., 2010; Bar-Nur et al., 2011).
Tumoral reprogramming and induction of pluripotency: similarities
The role of transcription factors in the control of tumoral reprogramming and induction of pluripotency
We have seen in the initial section of this review that both cancer research and developmental biology have been the focus of intense attention since ancient times. What's more, they have always been closely related from the conceptual point of view. The cellular theory of Rudolf Virchow is clearly essential for the understanding of both development and tumorigenenesis. But he went further, since he already proposed the embryonal rest hypothesis of tumour origin, after realising the histological similarities between tumours and embryonic tissues (Virchow, 1855). This concept was afterwards expanded by Julius Conheim, who suggested that tumours arise from residual embryonic remnants "lost" during normal development (Cohnheim, 1867). This hypothesis actually connects with the current theory of the cancer stem cells (CSCs) in which progenitors are situated at the root of cancer maintenance (see below). Another example of the influence of cancer research in the progress of the fields of stem cell biology and developmental biology is the fact that embryonic stem (ES) cells were identified in a search that has been initiated in the study of teratocarcinomas (Solter, 2006; Morange, 2007; Hochedlinger and Plath, 2009). In the field of cancer research it has traditionally been postulated that more than one molecular hit is required to generate a tumour cell, because several aspects of cellular biology must be altered in the progress towards a full-blown tumour (Hanahan and Weinberg, 2000). Therefore, in order to achieve tumoral reprogramming (although this was not the terminology traditionally used), more than one single molecular alteration had to happen. We have mentioned before that for a "simple" transformation, like a lineage switch, the change in the levels of expression of a single transcription factor could be enough (Davis et al., 1987; Nutt et al., 1999; Xie et al., 2004; Cobaleda et al., 2007a). Similarly, a single initial oncogenic lesion may contribute to just a part of the tumoral phenotype, by causing a block in differentiation, or an alteration in the control of cell cycle. In oncogenesis, many factors and routes have been shown to be altered, and their individual contributions to the tumoral phenotype are clear, although their synergy and interactions are less known. In the case of reprogramming to pluripotency, the discovery of Takahashi and Yamanaka (Takahashi and Yamanaka, 2006) revealed the nature of these factors. Before, reprogramming to pluripotency was only possible by the use of nuclear transplantation, but it was not known which of the factors present in the zygote possessed the required reprogramming capacity. Interestingly enough, the 4 Yamanaka factors are known to be involved in tumorigenesis in different contexts, and both c-Myc and Klf4 are well-known oncogenes (Rowland et al., 2005; Okita et al., 2007; Tanaka et al., 2007; Chen et al., 2008), thus further linking reprogramming to tumorigenesis. In summary, the experimental results show that the maintenance of cellular identity is essential for differentiated cells, and that only strong transcriptional or epigenetic regulators can subvert it. In this way, the multistep nature of tumorigenesis is paralleled by reprogramming to pluripotency in the series of "uphill" steps required and in the need for the sum of the effects of several factors to overcome the built-in safety mechanisms designed to protect cells from transformation or, in other words, to prevent cells from changing their identity. In the case of the reprogramming factors, the precise role of each of them is not yet clear, but their experimental introduction at different times during the process of reprogramming is shedding some light on this issue (Sridharan et al., 2009), by identifying distinct contributions of the different factors along the reprogramming progression. In the early stages of reprogramming, the most important process happening is the silencing of the gene expression programs of the differentiated cells. This aspect is previous to the induction of the ES-like expression program, and the main molecular responsible for this function seems to be c-Myc. However, it has also been shown that treatment with histone deacetylase inhibitors like valproic acid (VPA) can substitute for c-Myc, because of their capacity for repressing the gene expression programs of differentiated cells (Huangfu et al., 2008). Therefore, it would seem that the action of c-Myc takes place mainly before the activation of the regulators of the pluripotent state and, consequently, ectopic expression of c-Myc is required only during the first few days of the reprogramming process (Sridharan et al., 2009). In fact, c-Myc is dispensable for reprogramming, but in its absence there is an enormous drop in the efficiency of the procedure (Nakagawa et al., 2008; Wernig et al., 2008). The other three factors, Oct4, Sox2, and Klf4, need to act together to achieve the entry into the pluripotent condition, as evidenced by the fact that, when they are used individually, they cannot bind their pluripotent target genes in cells that are sill incompletely reprogrammed, most likely because the pattern of epigenetic modifications at these loci is not permissive for their binding (Sridharan et al., 2009). Indeed, Oct4, Sox2, and also Nanog co-bind to a plethora of genes in overlapping genomic sites (Boyer et al., 2005; Loh et al., 2006), in such a way that the transcriptional program required for pluripotency is maintained by the coordinated action of these key genes. In general, for the reprogramming of almost every cell type to pluripotency, the 4 Yamanaka transcription factors are enough. However, there are some exceptional cases in which additional alterations are required. For example, in the case of mature B cells it is necessary to interfere with the activity of the transcription factor Pax5, which is the master regulator of B cell identity (Cobaleda et al., 2007a; Hanna et al., 2008). Previous experiments had revealed that the elimination of Pax5, in the absence of any other genetic manipulation, allowed mature B cells to dedifferentiate to early haematopoietic multipotential progenitors (Cobaleda et al., 2007b). These findings again correlate reprogramming with cancer development, since it has also been shown that the elimination of Pax5 function in mature B cells induces a process of pathological dedifferentiation that gives rise to progenitor cell lymphomas (Cobaleda et al., 2007a). Therefore, the loss of a transcription factor that is required for the maintenance of cellular identity can be a tumour-inducing lesion. However, and contrary to mature B cells, earlier stages of B cell development can be reprogrammed to pluripotency in the presence of functional Pax5, just with the 4 Yamanaka transcription factors (Hanna et al., 2008), thus supporting the intuitive idea that the degree of differentiation of the target cell has an effect on the final efficiency of reprogramming (see below). In the genetic landscape, the oncogenic mutations alter the architecture of the whole gene regulatory network, since it modifies one of the nodes. This leads to an alteration in the landscape that gives rise to new abnormal attractors (new "valleys") where cancer cells reside (Huang et al., 2009). Furthermore, this alteration in the landscape gives the cell a new momentum to move towards new directions, and this effect can persist even when the initial stimulus has disappeared. From the point of view of tumoral reprogramming, this implies that the expression of a tumour-promoting gene, even if it is transient, can by itself trigger a durable malignant phenotype that does not require anymore of the initial mutation for its maintenance (Huang et al., 2009).
The role of epigenetic factors in the control of tumoral reprogramming and induction of pluripotency
In the previous section we have seen that either the gain or the loss of function of transcription factors plays an essential role in reprogramming to pluripotency, in the same way as how oncogene overexpression or loss of tumour suppressors promote tumorigenesis. Also, similarly to tumour progression, large-scale epigenetic changes are required for full reprogramming to happen. Today, it is clearly established that not only genetic alterations are responsible for cancer development, but there is also an important role of epigenetic alterations (Esteller and Herman, 2002; Esteller, 2007; Esteller, 2008) that lead to the specification of an heritable, abnormal pattern of gene expression that plays an essential role in cancer initiation and progression (Ting et al., 2006). All the relevant epigenetic marks, from DNA methylation to histone modifications, are perturbed in tumour progression. The subsequent changes in gene expression patterns are especially relevant when they affect the levels of expression of specific oncogenes or tumour suppressors, but they affect in fact the whole epigenome, and therefore condition all cellular identity. All these epigenetic alterations are usually secondary, and they can be just due to tumour progression and therefore independent from (i.e., not directly caused by) the initiating oncogenic mutation, but they can also be directly induced by the first oncogenic event, like it happens when chromosomal aberrations deregulate histone modification genes (Esteller, 2008). In the process of reprogramming to pluripotency, epigenetic modifications are intrinsically required for the process to take place, and they have to occur all throughout the genome, not being just restricted to the activation or repression of individual genes, something that is already achieved by the transcription factors. This explains why the efficiency of reprogramming is significantly superior in the presence of chemicals that can globally interfere with epigenetic marks. For example, the DNA methyltransferase inhibitor 5-aza-cytidine (AZA) causes a rapid and stable transition to a fully reprogrammed iPS state (Huangfu et al., 2008; Mikkelsen et al., 2008). Similarly, treatment with valproic acid (VPA), a histone deacetylase (HDAC) inhibitor, considerably improves the induction to pluripotency (Huangfu et al., 2008). Other example is provided by the use of the compound BIX-01294, an inhibitor of G9a methyltransferase that makes it possible to achieve reprogramming to pluripotency using only Oct4 and Klf4 transcription factors, with an efficiency comparable to the one obtained when using the four factors (Shi et al., 2008). In normal development, the biological role of G9a is to terminate the pluripotencial state as the progenitors exit to the differentiation process (Feldman et al., 2006; Epsztejn-Litman et al., 2008). This is achieved by its histone methylation activity, that prevents the reactivation of its target genes (for example embryonic genes like Oct4) when their transcriptional repressors are no longer present (Feldman et al., 2006). Also, at the same time, G9a promotes DNA methylation, that stops reversion towards the undifferentiated state (Feldman et al., 2006; Epsztejn-Litman et al., 2008). Therefore, genome-wide epigenetic changes affecting many still unknown loci, are essential in the late stages of direct reprogramming, and inhibition of the proteins responsible for generating or maintaining these marks lowers the "activation energy" required for the transition to pluripotency. Therefore, it makes sense that several of the chemical inhibitors that we have just mentioned are in fact already in use, or in clinical trials to be used as therapeutic agents against cancer. AZA was approved by the FDA in 2004 for the treatment of myelodysplastic syndromes, being the first drug into the new class of demethylating agents (Kaminskas et al., 2005). Its mechanism of action is very unspecific, aimed at the restoration of the normal levels of expression of genes whose expression has been lost due to promoter hypermethylation during tumoral progression, and that might be necessary for the control of proliferation and differentiation. Like in the case of most antitumoral drugs, AZA is expected to affect primarily the tumoral cells and leave non-proliferative cells unaffected (Sacchi et al., 1999; Kaminskas et al., 2005). Something similar happens for HDAC inhibitors (Dey, 2006; Lane and Chabner, 2009). All these findings underscore once more the concept of cancer as a reprogramming disease and a case of wrong differentiation.
Instructive and permissive factors in the progression and selection of the processes of tumoral reprogramming and induction of pluripotency
We have seen how both genome-wide changes in epigenetic marks and the loss and/or gain of transcriptional regulators are essential components of the processes of tumour generation and reprogramming to pluripotency. However, it is clear that these changes are clearly unwanted from the points of view of normal development and cellular function. Therefore, cells have developed many built-in protection mechanisms to maintain their identity against these transcriptional, genetic and epigenetic changes. Nevertheless, all these mechanisms are bypassed, in one way or another (Hanahan and Weinberg, 2000; Hanahan and Weinberg, 2011), and cancer appears. How this happens in "progression to pluripotency" (in analogy to tumoral progression) is still to be discovered. However, it has recently been shown by several groups (Zhao et al., 2008; Banito et al., 2009; Hong et al., 2009; Kawamura et al., 2009; Krizhanovsky and Lowe, 2009; Li et al., 2009; Marión et al., 2009; Utikal et al., 2009) that, exactly as it happens in cancer progression, the elimination of the DNA damage control checkpoint (p53-p21) greatly improves the efficiency of the reprogramming process, making it possible that many of the starting cells become successfully reprogrammed. This is done at the expense of an increased level of genetic instability, and most of the iPSCs obtained in the absence of a functional p53-p21 axis carry genetic aberrations of different kinds. This is in connection with what we have mentioned before about reprogramming being an "uphill", unfavourable process, which most of the cells fail to complete (Mikkelsen et al., 2008). Therefore, eliminating the DNA damage checkpoint diminishes the selection and allows a larger number of cells to survive until pluripotency. These results support the idea of cancer as a disease of cellular differentiation and, furthermore, reinforce the idea that suggests that the driving forces behind the tumoral process are aberrantly expressed transcription factors, epigenetic regulators and signalling molecules, while the role of many of the other alterations found in tumours (for example, the loss of p53) is mainly permissive.
Role of the cell of origin in tumoral reprogramming and induction of pluripotency
In the study of oncogenesis, it has traditionally been assumed that the phenotype of the tumour cells was a reflection of that of the normal cell that gave rise to the tumour in the first place. There were some classical examples in which this what not the case like, for example, chronic myelogenous leukaemia (CML), where the t(9;22) chromosomal translocation could be found in most types of differentiated haematopoietic cells, therefore indicating that a common, earlier progenitor, should be the cell of origin (Melo and Barnes, 2007). But, in general, since most cancerous cells are reminiscent of some differentiated cell type, for every type of tumour, the cell of origin was postulated to be the corresponding normal differentiated cell. However, the cancer stem cell (CSC) theory has led to a change in our perspective (Cobaleda and Sánchez-García, 2009; Vicente-Dueñas et al., 2009a; Vicente-Dueñas et al., 2009b). The CSC theory proposes that tumours are stem cell-based tissues just like any other, and this has several radical consequences for our understanding of cancer. The most important one is the fact that not all the tumoral cells are equally capable of regenerating the tumour. This means that, when tumoral cells are experimentally transplanted into a new host, or when some tumour cells remain in the patient after incomplete tumour excision, the reappearance of the tumour is caused by just a certain tumoral cellular subpopulation. Only those cells, possessing stem cell characteristics, can give rise to the whole tumour with all its cellular heterogeneity. Although there can be a big range of variability in the percentage of CSCs within a tumour, from very few to 25% (Quintana et al., 2008; Cobaleda and Sánchez-García, 2009; Vicente-Dueñas et al., 2009a; Vicente-Dueñas et al., 2009b), the fact is that, like in any other stem-cell based tissue, the majority of cells composing the tumour mass lack this capacity. Hence, if tumours are maintained by aberrant cells possessing stem cell characteristics, then what is the origin of these cells? This cancer cell-of-origin (not to be confused with the CSC, which would be the cancer-maintaining cell of the already developed tumour) is initially a normal cell (not necessarily a stem cell) that will be reprogrammed by the oncogenic events in order to finally originate (or convert into) a tumoral cell with stem properties. There are two main mechanisms that could be invoked in this scenario. One option is that the cell-of-origin suffering the oncogenic mutation(s) is already a stem cell, which therefore becomes reprogrammed to give rise to a new pathological tissue instead of the normal one. In the case of CML, it has recently been demonstrated, using genetically modified mice, that the restricted expression of the oncogenic alteration in the stem cell/progenitor compartment is enough to generate a human-like tumour with all the variety of differentiated tumour cells (Pérez-Caro et al., 2009; Vicente-Dueñas et al., 2009b). In mouse models of intestinal cancer it has also been found that tumours originate in the crypt stem cell, since when the oncogenic stimulus (activation of the Wnt signalling pathway) is targeted to the stem cell compartment, intestinal adenomas develop in which a developmental hierarchy is maintained. On the contrary, when the oncogenic lesions are targeted at the non-stem intestinal epithelial cells, they only generate short-lived, small microadenomas (Barker et al., 2008; Zhu et al., 2008). In the nervous system, targeting astrocytoma-associated oncogenic lesions to progenitors (in this case in the subventricular zone) results in tumour development, while targeting them to the differentiated cells of the adult parenchyma does not result in tumours, only in local astrogliosis (Alcantara Llaguno et al., 2009). Therefore, there are many examples (Dirks, 2008; Joseph et al., 2008; Zheng et al., 2008) where it has been proven that the initiating event takes place in a normal stem cell, even if the mature tumour is composed by differentiated cells, indicating a true tumoral reprogramming mediated by the oncogenic lesions (Vicente-Dueñas et al., 2009b). The other alternative is that the cancer cell-of-origin can be a differentiated cell that regains stem cell characteristics in the process of tumoral reprogramming. This option relies on two requirements: first, the oncogenic alteration must be capable of conferring or programming these characteristics in the target cell and, second, the cell must be plastic enough so as to be reprogrammed by this precise oncogenic alteration. It has been shown that some oncogenes, like MOZ-TIF2 (Huntly et al., 2004), MLL-AF9 (Krivtsov et al., 2006; Somervaille and Cleary, 2006), MLL-ENL (Cozzio et al., 2003), MLL-GAS (So et al., 2003) or PML-RARα (Guibal et al., 2009; Wojiski et al., 2009) can generate CSCs when they are introduced into committed target cells. Gene expression arrays have revealed that MLL-AF9 can activate a stem cell-like program in committed granulocyte-macrophage progenitors, therefore conferring them the property of self-renewal (Krivtsov et al., 2006). Also c-Myc can induce a transcriptional program reminiscent of that of embryonic stem cells in differentiated epithelial cells, and originate epithelial CSCs (Wong et al., 2008). However, other oncogenes are unable of conferring self-renewal properties, like for example BCR-ABLp190 (Huntly et al., 2004). In these cases the oncogene, since it cannot immediately confer stem cell properties, could give rise to a precancerous cell that can afterwards, with the presence of additional alterations conferring "stemness", give rise to the cancer stem cell (Chen et al., 2007). In any case, the cellular origin where the cancer-initiating lesions take place is difficult to determine since, in many cases, the functional impact of the oncogenic lesion (i.e. the tumour clonal expansion) can present with phenotypes mimicking differentiation stages that can be either upstream or downstream of the initiating cell. For example, the translocations that are the initiating lesions of many childhood B acute lymphoblastic leukaemias (ALL) originate in utero during embryonic haematopoiesis and promote the conversion of partially committed cells into preleukaemic cells with altered self-renewal and survival properties, that will require a second postnatal hit to develop into full leukemias (Hong et al., 2008). Also, in leukemias carrying the AML1-ETO translocation, this aberration can be detected in stem cells in patients in remission. These stem cells behave apparently normal during the remission phase, indicating that they can remain dormant and, with time, some of their descendants can become tumorigenic and originate the relapse (Miyamoto et al., 2000). We have described previously that, in mice, the loss of Pax5 in mature B cells leads to the dedifferentiation to multipotent progenitors and the appearance of progenitor B cell lymphomas (Cobaleda et al., 2007a). In human Hodking lymphomas, the overexpression of specific antagonists leads to the functional inactivation of the B cell factor E2A, which in turn causes the loss of B cell markers and induces the expression of lineage-inappropriate genes characteristic of the Reed-Sternberg Hodking lymphoma cells (Mathas et al., 2006). Also in children's B-ALLs, the CSCs can present with the phenotypes of different stages of early B cell development that, on top of that, can apparently interconvert among them, therefore complicating even more the task of identifying the cancer-cell of origin (le Viseur et al., 2008). A genomic analysis of samples from relapsed ALL patients, when compared with the samples at diagnosis, has shown that the same ancestral clone can be found at both stages of the disease (Mullighan et al., 2008). So, clearly in many cases the cancer-maintaining cell evolves over time and adapts to treatment to finally lead to relapse, and therefore the characteristics of the CSC population in a certain moment may not relate at all any more to those of the initial cancer cell-of-origin (Barabé et al., 2007). As we already mentioned when we described the view of reprogramming to pluripotency from the perspective of the GRNs, the inducing factors are not required anymore once the cells have reached the pluripotent condition and the new identity (however plastic this is) has been established. If cancer stem cells are generated by a tumoral reprogramming process, then maybe the oncogenes that initiate tumour formation might be not be required for tumour progression (Krizhanovsky and Lowe, 2009). If this were the case, it would explain the aforementioned examples in which a pre-cancerous lesion exists stably in an aberrant cell population that does only evolve to an open tumour when secondary mutations occur. In this scenario, the initiating lesion would be the driving force in the reprogramming process, but once this has been completed, it would only be a passenger mutation, or could even perform a different role that would be independent from its reprogramming capacity, like for example in tumour expansion/proliferation. A mechanism of this kind would explain why some targeted therapies fail in spite of their initial apparent efficacy: for example, imatinib, a drug targeted against the deregulated kinase activity of BCR-ABL, successfully eliminates differentiated tumour cells, but it fails to kill the BCR-ABL+ CSCs, since it does not seem to interfere with the function of the chimeric oncogene in this cellular context (Graham et al., 2002; Barnes and Melo, 2006). The fact that CSCs can originate from differentiated cells represents the last and most patent similarity between tumorigenesis and reprogramming to pluripotency. Also in iPSCs generation, the nature of the cell of origin is key in determining the global success. In this way, it has been described that, in the haematopoietic system, the capacity of reprogramming cells decreases as they differentiate, since HSC are 300 times more likely to be reprogrammed than B or T cells (Eminli et al., 2009). In the case of the nervous system, when the starting cells are adult neural stem cells (NSCs), then pluripotency can be achieved using only Oct4 (Kim et al., 2009), probably because of the high similarity of NSCs transcriptional profile to that of ES cells. Similarly, in a liver model of transdetermination it has been demonstrated that Neurogenin3 can convert hepatic progenitor cells into neo-islets but it cannot transdifferentiate mature hepatocytes (Yechoor et al., 2009).
Outlook
The knowledge obtained in the research of the molecular and cellular mechanisms that control cellular plasticity, pluripotency and reprogramming will also have a profound impact in our understanding of tumorigenesis and, in a more distant future, in the treatment of cancer. It is clear that the two fields of research will continue being mutually interdependent. By way of example, the main obstacle for the future use of iPSCs in the clinic is precisely the generation of tumours as a result of uncontrolled growth or differentiation of the cells, once they are in the patient. Therefore, the knowledge and control of the narrow limits of gene expression that mark the difference between normal and tumoral differentiation and reprogramming will be required before this problem can be overcome. Assuming the role that reprogramming plays in cancer generation makes it possible to initiate the development of new therapeutic strategies aimed at re-directing the wrong differentiation program towards a new outcome (ideally, in most cases, terminal non-tumoral differentiation and cellular death). Differentiation therapies are already in use in some cases, like the administration of retinoic acid to differentiate tumoral cells in PML-RARα+ positive acute promyelocytic leukemias. We have described how reprogramming to pluripotency, due to its inefficiency, can get caught up at several points before reaching the iPSC state (Mikkelsen et al., 2008). Tumoral cells are probably very close to these incompletely reprogrammed intermediates, and the study of the latter should help us in understanding how to get the former ones out of their pathologic block. In fact, epigenetic therapies are most probably going to be on the rise in the coming years for the treatment of many types of tumours, since our knowledge about the molecular mechanisms controlling the epigenetic marks and their role in self-renewal, differentiation and maintenance is increasing very quickly, and this should help us to obtain more and better (more specific) epigenetic drugs (Jones, 2007; Shen et al., 2009). The discovery of reprogramming to pluripotency has transfigured the research in the field of cellular plasticity. It is nowadays possible, using just three ectopic factors, to reprogram fibroblasts into functional neurons (Vierbuchen et al., 2010), to convert in vivo pancreatic exocrine cells to β cells (Zhou et al., 2008) or to directly transdifferentiate mouse mesoderm into heart tissue (Takeuchi and Bruneau, 2009). One of the most remarkable examples in this context is the phenotype caused by the deletion of a single gene, Foxl2, in adult ovarian follicles. This inactivation immediately upregulates testis-specific genes and leads to a full organ reprogramming (Uhlenhaut et al., 2009) that shows that the maintenance of the identity of the ovarian cells requires the active and constant presence of a specific gene. This is therefore an active process that resembles very much what we have described for Pax5 and B cells, but affecting a whole organ with all its cellular diversity. Our increasing knowledge and technical control over cellular identity should help us in the development of strategies for the reprogramming of tumoral cells. In fact, several experimental evidences seem to suggest that this is perfectly possible. For example, melanoma cells can be reprogrammed by nuclear transplantation (Hochedlinger et al., 2004). Also, embryonal carcinoma cells or mouse brain tumours have been used as a valid starting material for nuclear cloning experiments (Li et al., 2003; Blelloch et al., 2004). Therefore, maybe in a not so distant future we might have the knowledge and tools to manipulate tumoral cell identity to force cancer cells to differentiate, or to make them vulnerable to therapy.
Acknowledgments
Research in C.C. lab was partially supported by FEDER, Fondo de Investigaciones Sanitarias (PI080164), CSIC P.I.E. 200920I055 and 201120E060, from the ARIMMORA project (FP7-ENV-2011, European Union Seventh Framework Program) and from an institutional grant from the "Fundación Ramón Areces". Research in ISG group was partially supported by FEDER and by MICINN (SAF2009-08803 to ISG), by Junta de Castilla y León (REF. CSI007A11-2 and Proyecto Biomedicina 2009-2010), by MEC OncoBIO Consolider-Ingenio 2010 (Ref. CSD2007-0017), by NIH grant (R01 CA109335-04A1), by Sandra Ibarra Foundation, by Group of Excellence Grant (GR15) from Junta de Castilla y Leon, and the ARIMMORA project (FP7-ENV-2011, European Union Seventh Framework Program) and by "Proyecto en Red de Investigación en Celulas Madre Tumorales en Cancer de Mama", supported by Obra Social Kutxa y Conserjería de Sanidad de la Junta de Castilla y León. All Spanish funding is co-sponsored by the European Union FEDER program. ISG is an API lab of the EuroSyStem project. ECS is the recipient of a JAE-predoc Fellowship from CSIC and a "Residencia de Estudiantes" Fellowship. The authors declare no conflict of interest.